Ghost OS is an open-source framework that gives AI agents full computer use on macOS — controlling real applications through native accessibility APIs instead of screenshots. With about 1,700 stars on GitHub and development continuing in 2026, it is built natively in Swift and works with MCP-compatible agent clients.
Most computer-use agents rely on screenshots and pixel clicking, which is slow and brittle. Ghost OS takes a different route: it drives macOS applications through the accessibility layer, so agents can read interface structure directly and act precisely, with no screenshots required. Its workflows are self-learning — successful task routines are captured as reusable “recipes” that make repeated automations faster and more reliable over time. The project is part of a small family of related open-source tools from the same author, and it requires macOS 14 or later. Agents connect over the Model Context Protocol, so Ghost OS can plug into existing agent setups rather than demanding its own ecosystem.
The repository includes examples and setup documentation for connecting popular MCP clients, so you can evaluate the accessibility-driven approach on a Mac you already own before committing your automation workflows to it.
Ghost OS is free and open source under the MIT license. There is no paid tier or hosted version; you run it on your own Mac, and any model costs come from the agent framework and LLM provider you connect to it.
Developers building desktop agents and automations for macOS, and tinkerers who want AI assistants that can operate real applications reliably rather than guessing at pixels on a screen. It is a foundation for agent builders, not a consumer-facing app.